四川轻化工大学学报（自然科学版）2025，Vol.38Issue(6)：90-99,10.DOI:10.11863/j.suse.2025.06.10
基于YOLO-GELAN的无人机航拍图像目标检测方法研究
Research on a UAV Aerial Image Target Detection Method Based on YOLO-GELAN
摘要
Abstract
To address the issues of missed and false detections caused by large variations in target scale,dense distribution of small targets,and complex and variable backgrounds in UAV aerial imagery,an improved algorithm model(YOLO-GELAN)is proposed based on YOLOv8s.YOLO-GELAN enhances target key feature extraction and improves detection accuracy by introducing the RepNCSPELAN module from the Generalized Efficient Layer Aggregation Network(GELAN)to replace the original C2f module,while reducing network parameters.The neck network is then modified by adding a small target detection layer,removing the large target detection layer,and increasing the size of the feature map to be detected,thus enhancing the network's sensitivity to small targets.Finally,the Wise-IoU loss function is introduced to reduce large gradients caused by extreme samples,improving the overall performance of the network model.On the VisDrone2019 dataset,the algorithm has achieved precision,recall,and mAP50 of 55.7%,45.3%,and 47.0%,respectively,which represent improvements of 6.7%,7.9%,and 8.8%compared to the baseline network,with a 30%reduction in parameters.Experiments show that the improved model demonstrates superior performance in detecting small targets in images,making it effective for various UAV aerial imagery target detection scenes.关键词
YOLOv8/无人机航拍/GELAN/小目标层/损失函数Key words
YOLOv8/UAV aerial imagery/GELAN/small object layer/loss function分类
信息技术与安全科学引用本文复制引用
苏春亮,何小利..基于YOLO-GELAN的无人机航拍图像目标检测方法研究[J].四川轻化工大学学报（自然科学版）,2025,38(6):90-99,10.基金项目
2024教育部第三期供需对接就业育人项目(2023122021680) （2023122021680）
自贡市重点科技计划项目(自贡市医学大数据与人工智能研究院)(2022ZD16) （自贡市医学大数据与人工智能研究院）
企业信息化与物联网测控技术四川省高校重点实验室基金项目(2022WYY02) （2022WYY02）
四川轻化工大学教学改革研究项目(JG-24025) （JG-24025）
四川轻化工大学研究生创新基金项目(Y2024128) （Y2024128）
